Advanced Autonomous Systems: AI Reasoning in Physical Robots

Most robotics courses stop at getting the robot to move. This one starts there.

The problem this course addresses

Structured environments, like factory floors with fixed layouts, are a solved problem for most industrial robots. The hard part is everything else: construction sites, hospital corridors, outdoor terrain, shared spaces with unpredictable humans. Building AI systems that handle this requires a different set of tools and a different way of thinking about failure.

This program focuses on that harder problem. Students work through reinforcement learning for contact-rich manipulation, multi-agent coordination, and uncertainty-aware planning. Each module is built around a scenario that does not have a clean textbook answer.

Language models as reasoning engines

One module examines how systems like claude and claude AI can serve as high-level reasoning components in a robot architecture. Rather than treating the language model as a chatbot, we use it as a planner that interprets task descriptions, breaks them into subtasks, and monitors execution. Students implement this pattern and then stress-test it by deliberately introducing ambiguous instructions and partial failures to see where the reasoning breaks down.

Prerequisites and pace

This is a demanding program. Applicants should have prior experience with Python, some familiarity with machine learning concepts, and ideally have completed an introductory robotics course or equivalent work experience. Sessions run twice weekly and include substantial independent project time between meetings.

What you build

The final project requires deploying an autonomous task agent in a simulated unstructured environment. The agent must handle at least three categories of unexpected events without human intervention. Past participants have built systems for warehouse exception handling, assistive robot navigation, and multi-robot task allocation.